Teacher data generation method for three dimensional shape

Abstract

To provide a three dimensional shape teacher data generation method for achieving a generation AI technology related to a 3DCAD for highly accurately understanding a complicated 3DCAD and generating a three dimensional model corresponding to a precise instruction.SOLUTION: The method for generating teacher data for learning the three dimensional shape of a recognition object includes a step in which a 3DCAD data-acquiring part 2 acquires CAD (computer aided design) data of the three dimensional shape, and a step in which a linguistic expression-converting part 3 converts the CAD data of the three dimensional shape acquired by the 3DCAD data-acquiring part 2 into a STEPAP242 to generate teacher data of the three dimensional shape. As a result, it is possible to realize a generation and AI technique related to a 3DCAD capable of understanding even a complicated 3DCAD with high accuracy and generating a three dimensional model according to a precise instruction.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method for generating training data for three-dimensional shapes, and more particularly to a method for accurately learning the shape of an object (object to be recognized) having a three-dimensional shape. [Background technology]

[0002] Various methods for constructing three-dimensional models (3D models) have been proposed in the past. For example, Patent Document 1 discloses a CAD (Computer Aided Design) system for the purpose of speeding up updates of the three-dimensional model, the system including: a step of generating a three-dimensional model; a step of making changes to one of a plurality of components; a step of determining other components from the plurality of components that may change the three-dimensional model as a result of making the changes; and a step of updating the three-dimensional model by regenerating the other components determined as having the possibility of changing the three-dimensional model while not regenerating the remaining components that are not likely to change the three-dimensional model. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. WO2008 / 127254 Summary of the Invention [Problem to be solved by the invention]

[0004] While conventional generative AI (Artificial Intelligence) technology can create text-based summaries and reports, generate various images and videos, and perform programming, there is currently little generative AI technology related to three-dimensional shapes.

[0005] In other words, AI learning of three-dimensional shapes (3D solid shapes) extracts features and patterns from visual information such as images and videos, and identifies the three-dimensional shape based on those features and patterns. Therefore, it is possible to recognize the three-dimensional shape from an image, determine its type, and determine whether it has any unusual features.

[0006] However, there is ambiguity in AI's recognition of 3D shapes, and it has not yet been possible to recognize 3D shapes with high accuracy. As a result, it has not yet been possible to realize generative AI technology related to 3D CAD that can understand complex 3D CAD data with high accuracy or generate 3D models according to precise instructions.

[0007] The present invention has been made in consideration of these points, and its purpose is to provide a method for generating training data for 3D shapes that enables the realization of 3D CAD-related generation AI technology that can understand complex 3D CAD data with high accuracy and generate 3D models according to precise instructions. [Means for solving the problem]

[0008] The solution of the present invention for achieving the above object is based on a method for generating training data for learning the three-dimensional shape of a recognition target, which includes the steps of acquiring CAD data of the three-dimensional shape, and converting the acquired CAD data of the three-dimensional shape into a predetermined protocol to generate training data of the three-dimensional shape.

[0009] This specification allows the AI ​​to learn the 3D shape of the object to be recognized based on the training data generated by this converted protocol by converting CAD data into a specified protocol. This makes it possible to realize generative AI technology related to 3D CAD that can understand even complex 3D CAD data with high accuracy and generate 3D models according to precise instructions.

[0010] Specifically, the conversion converts the CAD data of the three-dimensional shape into a protocol constructed by linguistic expression.

[0011] Because the 3D shape is identified by a protocol constructed using linguistic expressions, the AI ​​can learn the 3D shape of the object to be recognized with high accuracy (language learning), making it possible to realize generative AI technology related to 3D CAD that can understand even complex 3D CAD data with high accuracy and generate 3D models according to precise instructions.

[0012] Furthermore, the three-dimensional shape of the object to be recognized is composed of a plurality of parts, and when converting the CAD data of the three-dimensional shape into a predetermined protocol, a step of reflecting the definitions of the plurality of parts is added.

[0013] This will enable the realization of generative AI technology related to 3D CAD that can understand with high accuracy even the complex 3D shapes of objects to be recognized that are made up of multiple parts, and can generate 3D models according to precise instructions.

[0014] The definition of the plurality of regions is achieved by annotating each of the regions.

[0015] This allows the AI ​​to give precise instructions corresponding to each part when executing a prompt for that part. [Effects of the Invention]

[0016] In the present invention, when generating training data for learning the 3D shape of a recognition target, the steps of acquiring CAD data of the 3D shape and converting the acquired CAD data of the 3D shape into a predetermined protocol to generate training data of the 3D shape are carried out. This makes it possible to realize a generative AI technology related to 3D CAD that can understand even complex 3D CAD data with high accuracy and generate a 3D model according to precise instructions. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a block diagram showing an outline of a three-dimensional shape recognition support system according to an embodiment. [Figure 2] FIG. 2 is a diagram for explaining a learning process of 3D CAD data in an embodiment. [Figure 3] 10A to 10C are diagrams for explaining a process of linking a three-dimensional shape with a part in an embodiment. [Figure 4] FIG. 10 is a diagram for explaining a 3D model generation process when a prompt is executed in an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0018] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In this embodiment, a functional unit that implements the method for generating training data for a three-dimensional shape according to the present invention will be described as being provided in a three-dimensional shape recognition support system, which will be described later. However, the configuration for implementing the method for generating training data for a three-dimensional shape is not limited to this.

[0019] -Outline of the 3D shape recognition support system- FIG. 1 is a block diagram showing an outline of a three-dimensional shape recognition support system 1 having a functional unit for implementing a method for generating training data for a three-dimensional shape according to the present invention.

[0020] 1, the three-dimensional shape recognition support system 1 according to this embodiment includes a 3D CAD data acquisition unit 2, a language expression conversion unit 3, a machine learning unit (AI) 4, a CAD system unit 5, and a prompt input unit 6. The machine learning unit 4 includes a language information learning unit 41 and a prompt execution unit 42. However, the configuration of the three-dimensional shape recognition support system 1 is not limited to this.

[0021] (3D CAD data acquisition section) The 3D CAD data acquisition unit 2 is a functional unit that acquires 3D CAD data of a recognition target object (e.g., an automobile body) for which high-precision three-dimensional shape recognition is required. This 3D CAD data is, for example, data created by an external device or the like, and is input from the external device to the three-dimensional shape recognition support system 1 by an operator. The operation of acquiring 3D CAD data in this 3D CAD data acquisition unit 2 corresponds to the "step of acquiring CAD data of a three-dimensional shape" in this invention.

[0022] (Language Expression Conversion Unit) The language expression conversion unit 3 is a functional unit that receives the 3D CAD data acquired by the 3D CAD data acquisition unit 2 and converts this 3D CAD data into a language expression of STEP (Standard for the Exchange of Product Data) AP242, which is an ISO standard.

[0023] Specifically, the EXPRESS modeling language is used as this language representation. The EXPRESS modeling language is a formal specification description language prepared for formally describing product model information in STEP (an international standard for the external representation of product models being deliberated by Subcommittee 4 of the ISO Technical Committee 184). The language representation may also include 3D model data saved in ASCII text format.

[0024] The information (3D language information) of STEPAP242 converted from the 3D CAD data in this way is used as training data for learning the shape of the object to be recognized in the machine learning unit 4, which will be described later. Therefore, the conversion operation to a language expression in the language expression conversion unit 3 corresponds to "the step of converting the acquired CAD data of the three-dimensional shape into a predetermined protocol and generating training data of the three-dimensional shape" and "converting the CAD data of the three-dimensional shape into a protocol constructed by a language expression" in the present invention.

[0025] More specifically, STEPAP242 is a well-known standard for data exchange, also known as ISO10303, and is an ISO standard for image data exchange. STEPAP242 3D model data is constructed using text (linguistic expression) that can be interpreted by various CAD systems. Using STEPAP242 makes it easy to create, share, and edit 3D models using a variety of programs and software.

[0026] The reasons for using STEPAP242 as the conversion standard include its compatibility with a variety of CAD tools and software, making it easy to share and edit as mentioned above, and the use of NURBS curves, which mathematically represent curves, allowing data to be saved accurately, making it easy to customize, and making it easy to save on another computer as a backup.

[0027] In this embodiment, STEPAP242 is used as the standard to be converted to, but the present invention is not limited to this, and conversion to STEPAP203, STEPAP214, IGES (Initial Graphics Exchange Specification), or STL (Stereolithography, Standard Triangle Language) may also be used.

[0028] STEPAP203 is known to define geometry, topology, and configuration management data for solid models of parts and assemblies.

[0029] STEPAP214 is known to encompass all of the information contained in STEPAP203, plus information on color, layers, design intent, and geometric dimensional tolerances.

[0030] IGES is a file format for CAD intermediate data that conforms to the IGES standard, and is known as an ASCII text file consisting of 80 characters per line.

[0031] Furthermore, STL describes a set of small triangular elements that make up a three-dimensional shape, and is known to exist in an ASCII STL format described in plain text and a binary STL format.

[0032] In this embodiment, as described above, STEPAP242, STEPAP203, STEPAP214, IGES, and STL are given as examples of standards to which 3D CAD data can be converted. However, the present invention is not limited to these, and any standard can be applied as long as it can be converted into a language expression that makes it easy for AI to understand the three-dimensional shape.

[0033] (Language Information Learning Department) The linguistic information learning unit 41 of the machine learning unit 4 has the function of meticulously and accurately learning the shape of the object to be recognized based on the information of the given linguistic expression (information converted into the aforementioned STEPAP242). In other words, by recognizing 3D CAD data using information of linguistic expression that is easy for AI to understand as training data, the machine learning unit 4 is able to meticulously and accurately learn the shape associated with the 3D CAD data (the shape of the object to be recognized) with high accuracy, even for complex 3D CAD data. This eliminates the ambiguity in the previous recognition of 3D shapes by AI, making it possible to recognize the 3D shape of the object to be recognized with high accuracy.

[0034] (Prompt execution section) As will be described later, the prompt execution unit 42 of the machine learning unit 4 has a function of, when a prompt (instruction) is input from the prompt input unit 6, performing image processing according to the prompt on the recognition target object whose shape has been meticulously and accurately learned by the language information learning unit 41. Examples of this prompt include changing the outer shape of the recognition target object, changing the size of each part on the recognition target object, changing the position of each part on the recognition target object, changing the color of each part on the recognition target object, and so on.

[0035] (CAD Systems Department) One of the functions of the CAD system unit 5 is to define each part of the shape of the recognition object obtained by the learning process by using annotations. These annotations are also converted into linguistic expression information. The conversion into linguistic expression in this case is performed in the same manner as the conversion described above (for example, conversion into STEPAP242), so a detailed description is omitted here.

[0036] In this conversion, the conversion destination standard is not limited to STEPAP242, but may be STEPAP203, STEPAP214, IGES, or STL. Furthermore, it is not limited to these, and any standard can be applied as long as it can be converted into a linguistic expression that makes it easy for AI to understand the three-dimensional shape of each part of the shape of the object to be recognized.

[0037] As a result, the machine learning unit 4 (language information learning unit 41) learns by linking the shape of the recognition target object with each part (information such as the position, size, shape, etc. of each part). Therefore, the annotation using the functions of this CAD system unit 5 corresponds to "the step of reflecting the definitions of multiple parts when converting CAD data of a three-dimensional shape into a predetermined protocol" in this invention.

[0038] In this embodiment, the shape of the object to be recognized is linked to each part by annotations that utilize the functions of the CAD system unit 5, but the shape of the object to be recognized may also be linked to each part by other means.

[0039] (Prompt input section) The prompt input unit 6 is a functional unit that allows an operator to input instruction information such as changes to the shape of the recognition object or each part of the object. For example, if the recognition object has a hole, a prompt to change the position of the hole is input.

[0040] -Example of application to car body shape recognition- Next, an example of a specific usage of the 3D shape recognition support system 1 configured as described above will be described. Here, an example will be described in which an AI is made to learn 3D CAD data of a car body, and a 3D model is generated in which the position of a hole drawn on a part of the car body is moved a predetermined distance (a specified distance) toward the front of the car body as a prompt. The prompt is not limited to this, and various other instructions can be given, such as changing the size of each part on the car body, changing the position of each part on the recognition object, and changing the color of each part on the recognition object.

[0041] Figure 2 is a diagram explaining the learning process of 3D CAD data, Figure 3 is a diagram explaining the process of linking 3D shapes and parts (here, hole positions), and Figure 4 is a diagram explaining the 3D model generation process when a prompt is executed.

[0042] First, in the 3D CAD data learning process, as shown in Figure 2, 3D CAD data of the vehicle body V is acquired by the 3D CAD data acquisition unit 2 (see Figure 1), and this 3D CAD data is converted into STEPAP242 language expression (3D shape expression language) by the language expression conversion unit 3.

[0043] Then, the language information learning unit 41 of the machine learning unit 4 meticulously and accurately learns (language learning) the three-dimensional shape of the vehicle body V based on the 3D shape representation language STEPAP242.

[0044] Furthermore, in the process of linking the three-dimensional shape with the hole position, as shown in FIG. 3, the CAD system unit 5 uses its functions to define each part of the shape of the vehicle body V by annotation (information such as the position and shape of each part is verbalized by annotation), which allows the machine learning unit 4 to link the three-dimensional shape of the vehicle body V with each part with high precision. In particular, the position (position on each of the X-axis, Y-axis, and Z-axis) of the hole H1 formed on the front fender inner panel of the vehicle body V is linked with the shape of the vehicle body V with high precision.

[0045] When executing a prompt to change the position of hole H1, such as when making a design change, the prompt information (here, an instruction to move the hole position) is input to the AI, as shown in Figure 4. Figure 4 shows a case where prompt information is input to the AI ​​to move the position of a hole formed on the front fender inner panel of vehicle body V from position H1 in the figure (position indicated by the dashed line) to position H2 in the figure (position indicated by the solid line). The prompt information input to the AI ​​in this case includes information for identifying the hole (part in the 3D CAD data), the direction in which the hole should be moved (such as the movement distance on each of the X-axis, Y-axis, and Z-axis), etc.

[0046] The AI ​​that receives this prompt information will generate a 3D model to move the position of the hole on the front fender inner panel of vehicle body V from position H1 to position H2. As described above, the 3D shape of vehicle body V has been meticulously and accurately learned using the 3D shape representation language STEPAP242 as training data, and each part of the shape of vehicle body V has been defined by annotation, so that each part is linked with high precision to the shape of vehicle body V. Therefore, the position of hole H1 is recognized with high precision, and the amount of movement from that position according to the prompt information (the amount of movement from hole H1 to hole H2) can also be obtained with high precision. As a result, the generated 3D model faithfully reproduces the prompt.

[0047] -Effects of the embodiment- As described above, in this embodiment, acquired 3D CAD data is converted into STEPAP 242 to generate training data for the 3D shape of the object to be recognized. This allows the machine learning unit 4 to learn the 3D shape of the object to be recognized based on the training data generated by the converted STEPAP 242, thereby realizing a generative AI technology related to 3D CAD that can understand even complex 3D CAD data with high accuracy and generate a 3D model according to precise instructions. In particular, because STEPAP 242 is constructed using linguistic expressions, the machine learning unit 4 can learn the 3D shape of the object to be recognized with high accuracy.

[0048] In addition, in this embodiment, when converting 3D CAD data, a step is performed to reflect the definition of the parts of the object to be recognized. This makes it possible to realize a generative AI technology related to 3D CAD that can accurately understand even complex 3D shapes of objects to be recognized that are composed of multiple parts and generate 3D models according to precise instructions. In particular, since the definition of these multiple parts is performed by annotating each part, when executing a prompt for each part, it is possible to provide the machine learning unit 4 with accurate instructions corresponding to the part.

[0049] -Other embodiments- The present invention is not limited to the above-described embodiments, and all modifications and applications within the scope of the claims and equivalents thereto are possible.

[0050] For example, in the above embodiment, an example was described in which an AI is trained on 3D CAD data of a vehicle body V, and a 3D model is generated by shifting the position of the hole H1 in the vehicle body V by a predetermined distance toward the front of the vehicle body as a prompt. The present invention is not limited to this, and can also be applied to cases in which a 3D model is generated by a prompt targeting another part of the vehicle body V.

[0051] Furthermore, the target is not limited to 3D CAD data of the car body V, but can also be 3D CAD data of various objects (objects to be recognized).

[0052] For example, in the case of automatic generation of 3D models, it can be applied to automatic generation of 3D models of factory production lines, automatic generation of 3D models of buildings such as houses and other structures, and automatic generation of 3D models of electrical devices such as smartphones and televisions.

[0053] It can also be applied to the allocation of tasks related to 3D models. For example, it can be used to automate the allocation of assembly tasks by factory workers or production robots based on the 3D CAD data of a vehicle, or to automate the checking of regulations, production requirements, safety requirements, etc., based on the 3D CAD data of vehicles and production equipment. [Industrial Applicability]

[0054] The present invention is applicable to a method for generating training data for three-dimensional shapes that can be used for designing car bodies, etc. [Explanation of symbols]

[0055] 1. 3D shape recognition support system 2. 3D CAD data acquisition section 3 Language Expression Conversion Unit 4. Machine Learning Department 41 Language and Information Learning Department 42 Prompt Execution Unit 5 CAD Systems Department 6 Prompt input section V Car body (recognition target) H1, H2 holes (areas on the object to be recognized)

Claims

1. A method for generating training data for learning a three-dimensional shape of a recognition target, comprising: acquiring CAD data of the three-dimensional shape; a step of converting the acquired CAD data of the three-dimensional shape into a predetermined protocol and generating training data of the three-dimensional shape.

2. 2. The teacher data generation method according to claim 1, The training data generating method is characterized in that the conversion converts the CAD data of the three-dimensional shape into a protocol constructed by linguistic expression.

3. 3. The teacher data generation method according to claim 1, the three-dimensional shape of the object to be recognized is composed of a plurality of parts, A teaching data generating method characterized in that a step of reflecting definitions of the plurality of parts is added when converting the CAD data of the three-dimensional shape into a predetermined protocol.

4. 4. The teacher data generation method according to claim 3, A training data generation method characterized in that the definition of the multiple parts is achieved by annotating each of the parts.

Images